Is there a name for this specific family of rational approximations?
The general form of these series is that each term is a power of 10 (since moving the decimal takes no effort) multiplied or divided by a single-digit number (since single-digit multiplication/division takes far less effort than multi-digit multiplication/division).
It follows the basic principle of simple repeated fractions, but with only adding and subtracting error terms (no taking reciprocals and then cross-multiplying) and with having multiple options at each step from which to choose the best (rather than being given one automatically).
Taking π = 3.14159265, for example, we would start with either 3 (underestimating with 10^n times x), 4 (overestimating with 10^n times x), 10/4 (underestimating with 10^n divided by x), or 10/3 (overestimating with 10^n divided by x).
3 is the closest starting point (error of 0.14159265) and 3.33333333 is almost as close (error of 0.19170408), so we would throw 4 and 2.5 away and test 3 first, then 3.33333333.
The error “π – 3 = 0.14159265…” can be estimated as 1/10, as 2/10 (which could be re-written as 1/5, but that doesn’t matter here because we’re about to ignore it anyway), as 1/8, or as 1/7.
1/7 and 1/8 are the closest second-steps for π ≈ 3, so we throw away the 1/10 and the 2/10, and now we see what the closest second-steps would be for π ≈ 10/3.
The error “π – 3.33333333 = -0.1917408” would best be approximated as -0.2 (which could be written as -1/5 or -2/10 depending on the reader’s personal preference), but the two-step calculations 10/3 – 1/5 = 3.1333333333 and 3 + 1/8 = 3.125 are both less accurate than the two-step calculation 3 + 1/7 = 3.14285714, so we can commit to 3 + 1/7 now.
Calculating the new error “π – (3 + 1/7) = -0.00126449” creates a best new error term of -1/800, and so our new value is 3 + 1/7 – 1/800 = 3.14160714.
This approximation “πx ≈ 3x + x/7 – x/800” is accurate to within 1 part in 220,000, but it only takes about as much time and effort as “πx ≈ 3x + x/10 + 4x/100” (which is only accurate to within 1 part in 2,000).
Using continued fractions would take very little time to calculate “π ≈ 355/113” ahead of time (which is accurate to within 1 part in 12,000,000), but this takes more time and effort to use in the moment. If someone was multiplying πx ≈ 3x + x/7 – x/800 and if someone else was multiplying πx ≈ (300x + 50x + 5x)/113, then in the time it took the first person to get a final answer, the second person would only have finished calculating the numerator, and they would still need time to calculate the denominator.
During which time, the first person could be adding more error terms: 3 + 1/7 – 1/800 – 1/70,000 is accurate to within 1 part in 15,000,000 (already more accurate than 355/113 for less time and effort), and 3 + 1/7 – 1/800 – 1/70,000 – 1/5,000,000 is accurate to within 1 part in 880,000,000.